Reading AI answers

Use Answers, under an individual report’s AI tab, to review what each enabled AI platform returned for a tracked prompt, whether the visible answer names the client brand, and how that mention was portrayed when sentiment evidence is available.

This report is a captured result for one project, keyword, platform, location segment, and scan. It is not a statement about every response that platform can produce.

Before you start

The tracking project must include at least one AI platform and have a completed AI scan. Platform cards appear only for selected platforms with current, previous, empty, or failed result state.

Review supported tracking platforms before adding platforms or extra AI locations because each successful check adds workload.

Read an answer

1. Open the relevant tracking project and keyword report.
2. Choose AI, then Answers.
3. Select the primary or additional AI location when location segments are available.
4. Expand a platform card to read the captured answer.
5. Compare the capture date and historical answers before treating a wording change as a trend.
6. Open AI citations when you need the pages that supported the answer.

The answer shown inside a platform card and its Mentioned status refer to the same visible response. Rankability checks the displayed answer prose against the tracked brand and its configured aliases.

How an answer is laid out follows the structure the provider returned. Google AI Overview and AI Mode answers keep the provider’s text blocks, Perplexity answers keep their headings, lists, tables, and citation markers, and other platforms show the captured answer as formatted text. A plainer card means the provider returned less structure, not that Rankability captured less.

Interpret the platform badges

State	Meaning
Mentioned	The visible answer prose names the tracked brand or a configured alias.
Cited as source	A recognized client source is cited, but the visible answer does not name the brand.
Not mentioned	An answer was captured, but its visible prose does not name the brand.
No AI Overview / No AI answer	The platform check succeeded, but that surface did not return an AI answer for the query. This is not the same as a failed scan.
Scan failed	The platform did not complete successfully. Previous data may be shown and labeled as such.

Google AI Overview, Google AI Mode, and Brave AI do not generate an answer for every query. A neutral no-answer state means the surface was absent for that run; it does not mean an answer existed and omitted the brand. Rankability does not bill that unavailable surface when the product explicitly says no answer was returned.

Read sentiment carefully

When a visible answer mentions the brand and sentiment was captured, the card can show Positive, Neutral, Mixed, or Negative, with a short evidence reason.

Sentiment describes the tracked brand in that answer—not the overall tone of the entire response and not the reputation of the brand everywhere. A dash means sentiment was not captured for the answer; it should not be converted into a neutral classification.

Mentions and citations are separate

These states can occur independently:

• The answer names the brand but cites only third-party sources.
• A client page is cited as a source but the answer never names the brand.
• The answer both names the brand and cites a client page.
• The answer contains neither a brand mention nor a client citation.

SPI therefore calculates AI mentions and AI citations as separate categories. Where per-platform tracking-mode controls are available, a project can track both, citations only, or mentions only for an AI platform.

Compare brands in AI search rankings

AI search rankings compares every brand named across the AI answers for this report. Snapshot ranks them for the current scan with their platforms, visibility, average position, and score; Trend charts the same entities over time. Use Platforms to limit the comparison, and Show all when you want more than the leading rows.

A brand that belongs to the tracked brand’s own recorded identity is labeled as a tracked entity associated with the primary brand instead of being ranked as a competitor. The trend chart marks the primary brand (You) and those entities (Related), so a second business name of the same client is never read as a rival.

When related entities are present, use the scope chip in the AI score card header to show visibility for All tracked entities, the primary brand, or one related entity. Read the All tracked entities score carefully: it counts the primary brand and its associated entities together, each platform once, so it is not evidence that the primary brand itself appeared in every answer. The score card note says so whenever that scope is selected.

Which names count as the primary brand and which are separate related entities is set per client in Client settings → Brand identity.

Export CSV downloads every filtered row, not only the rows on screen, and keeps that distinction: Entity type records whether a row is the primary entity, a related tracked entity, or a competitor, and Associated with names the primary brand for a related entity. The file follows the platforms you have selected.

Location and history

Additional AI locations create separate answer segments for location-eligible model platforms. Google AI Overview and AI Mode use the search result’s own location context instead of the extra model-location fanout.

Use historical answers to compare the same platform and location over time. AI wording and citations can vary between runs, so confirm a repeated pattern before making a broad claim or changing strategy.

Review fan-out queries

Some providers expose the searches they used while researching an answer. When Rankability captures that evidence, the platform card shows a collapsed Fan-out queries section with the number of distinct queries returned.

Expand the section to review the query chips. Select one or more useful queries, then choose Add to keyword list to save them to an existing Researcher list or create a new list. Rankability recognizes queries already saved in that list rather than adding duplicates.

Fan-out evidence is currently captured when ChatGPT, Gemini, Claude, Grok, or Meta returns explicit query text. The section stays absent when the provider did not expose a query. Rankability does not infer a fan-out query from a citation, URL visit, search result, or the wording of the answer, and Perplexity results therefore do not show this section unless explicit query evidence becomes available.

Treat a fan-out query as evidence of the research path for that exact captured answer. It can suggest follow-up research or content coverage, but it does not prove that every response for the prompt will use the same searches.

You can also ask Serena to analyze saved fan-out evidence across the workspace’s active Tracker reports. Serena can identify queries that recur across tracked topics, platforms, and scans; show when each pattern was first and last captured; and use those supported patterns to suggest content priorities. This analysis reads existing scans only, so it does not rerun searches or use additional tracking allowance.

Serena’s saved fan-out analysis also includes an expandable View fan-out tree. Open it to see tracked keywords branch into the saved search queries, with platform names, capture counts, and capture dates. Capture details shows the first and last seen dates and up to two latest captures in UTC. These branches show observed keyword-to-query relationships, not the order of the AI’s reasoning or newly generated queries.

The tree is a bounded preview: up to 12 keywords and 12 queries per keyword, within the requested analysis window. Its counts and partial-preview label make omissions visible, including when the underlying saved-result limit is reached. It is not a complete historical export. Open Tracker report opens the canonical keyword report, not a date-pinned capture. The tree is saved with the Serena reply and remains available when you reopen that conversation; older replies made before this feature do not acquire a tree automatically.

ChatGPT local results and sponsored results

Some ChatGPT responses include structured results in addition to answer prose:

• ChatGPT local results lists the businesses captured with the answer, in the order they were returned, with the thumbnail, category or address, and rating Rankability captured for each one. Use Copy names to copy the listed business names, or download the captured details when you need the whole record.
• Sponsored results are shown separately from local businesses. Rankability only treats a result as an ad when the captured provider data contains usable sponsored creative; a local listing is not labeled as an advertiser merely because it has a business name or logo.

Select a business to open its captured detail: the rating and the source it came from, website, directions, and menu links, address, phone, the hours as captured, photos, a description, review excerpts and highlights, recommendations, and the sources cited for that business. A link to the original conversation and the capture time close the panel, and the panel says plainly when only part of the expanded detail was captured.

Records that were returned with the map but whose featured placement is not confirmed are grouped under a separate disclosure for additional map records, so they are never mixed into the ranked listings. When part of the capture could not be interpreted, the card says so and the download holds those details.

Structured fields depend on what ChatGPT returned for that scan. A missing rating, image, business name, review, or ad detail means that field was not available in the captured result; it should not be converted into zero, inferred from another result, or read as proof that ChatGPT did not show it.

Export complete answers

From the Tracker project list, select one or more reports and open the export menu. Full spreadsheet gives you a report worksheet for each selection, including a full-AI-answer section with the captured answer text and its keyword, platform, capture time, location, mention state, citation state, sentiment evidence, cited URLs, provider, and provider tier.

When the export covers more than one report, consolidated tabs are added in front of the per-keyword tabs, which stay as they are:

• Summary — brand mention and citation rates by platform and by keyword, share of voice, source coverage, and a link to each keyword tab.
• Citation master — one row per cited URL across every keyword, sorted by how many keywords cite it and then by total score, with the strongest mention status for that URL.
• Outreach targets — Fix the link for pages that name the brand in a link pointing somewhere else, and Get mentioned for outreach-ready pages that do not mention the brand and are cited for two or more keywords. Competitors and the client’s own site are left out.
• Domains — the same citations grouped by website.
• All citations — every keyword’s citation rows in one table for filtering.
• Read me — what each tab and column means.

The workbook reads the same labels the app shows, so a status never appears as a raw code and platforms are named rather than listed by identifier. Outreach fit carries the outreach classification — Opportunity, Competitor, Your site, or Not an outreach target — with Not a target because beside it, while the page kind stays under URL type. Answer text is plain text with its list structure kept, titles are cleaned, URLs drop tracking parameters, dates are real date cells, and a blank cell means nothing was recorded rather than a zero.

Fanout Queries is a narrower export: the unique queries captured across every saved run of the selected reports, each with the original prompts, platforms, first and last seen dates, and run counts. Set a From and Through date to limit it, or leave both blank for all history; dates are UTC. Only queries Rankability actually captured appear, so a run whose queries were not captured contributes nothing to the file.

Very long answers can continue across numbered rows because Excel limits the text stored in one cell. Join those continuation rows in order when analyzing the answer outside Rankability. A report with no captured answer does not receive a fabricated answer row.

The workbook is a historical evidence export. It does not rerun the platforms or refresh an answer at download time. Compare the capture timestamp before using it in a current client claim.

Failed and stale results

If a platform fails, Rankability can show previous data with a failure notice. Do not report that answer as newly captured. Use Retry where available and avoid parallel duplicate retries.

If no AI response data is available, run an AI scan rather than a Traditional or Video-only scan. Those narrower scans do not refresh AI answers.

Act on the result

• If the brand is absent, compare named competitors and the evidence they receive before deciding what to change.
• If the answer is inaccurate or unfavorable, preserve the exact answer and source evidence before planning remediation.
• If a third-party page supports the answer, review its mention status in AI citations and move qualified outreach targets into Prospector.
• If the client page needs improvement, continue with Page Content Auditor or Copywriter.

An absent mention is evidence for that measured prompt and scan—not proof that the brand never appears on the platform.